Embodied AI’s Air robot targets flexible factory work

By Nathan Brooks

Embodied AI unveiled Air on Nov. 6, 2026, at Humanoid Europe 2026 in Munich, nine months after starting development. The Swiss robotics startup says the wheeled humanoid learns manual factory jobs through human demonstrations, works alongside employees and typically shifts to autonomous operation within about a month.

Embodied AI says testing is already underway at major manufacturers. The company also says it has signed contracts with major manufacturers globally and is preparing Air for larger-scale production.

Embodied AI designed Air for repetitive work that demands precise hand movements and frequent task changes, conditions the company says conventional industrial robots struggle to automate. Rather than use traditional arms built for heavy industrial work, Embodied AI built lightweight, tendon-driven arms to reduce injury risk when the robot contacts people.

Embodied AI places Air’s motors, gearboxes and electronics in the torso. Tendons control the lightweight arms, which the company says yield when they meet an obstruction or touch a person. Air uses wheels rather than legs because Embodied AI says that arrangement provides greater stability on flat factory floors.

Embodied AI designed Air under ISO/TS 15066:2016, a technical specification addressing safety for collaborative industrial robots. The company plans CE marking for Air in 2027.

Learning from human demonstrations

Under Embodied AI’s learning model, Air initially performs each job under human control and records the activity. The company’s AI software learns the movements and eventually performs the work independently.

The transition typically takes about a month, depending on the task, according to Embodied AI. Human operators remain available for unfamiliar situations, and one robot can transfer learned skills to others.

What does Air change for factory operators?

For factories with frequently changing manual work, Air changes deployment from relying on a conventional industrial robot to showing the task through a human demonstration. Embodied AI says personnel guide the initial run, the software then performs the task independently, and trained operators remain available for exceptions. The practical change is a deployment model built around a roughly one-month learning period, not immediate autonomy.

Embodied AI says its AI software already operates in paid factory deployments through a separate two-armed industrial robot system. That system provides the company’s disclosed commercial performance benchmark.

In an evaluation by a major consumer electronics manufacturer, Embodied AI reported 98% reliability on a cable-handling task. The customer’s target was 90%. Embodied AI said four other startups participating in the evaluation failed to meet the manufacturer’s requirements.

The 98% figure applies to the two-armed system, not Air. Embodied AI is still testing the wheeled humanoid while preparing it for larger-scale production.

Co-founder and CEO Francesco Stella said safety and reliability were central to Air’s development because errors in physical tasks can cause damage or injuries. He said:

“Hallucination of AI models is acceptable in the digital space because humans review, reject, or correct the output before acting on it,” Stella said.

“In robotics, however, incorrect actions have immediate and irreversible physical consequences. When AI is used to taking physical actions, errors can cause severe damage or endanger human lives. Safety and reliability are non-negotiable requirements of physical AI to enter the real world at scale. These are the principles that guided the design of Air.”

Air’s tendon-driven arm design grew out of Stella’s doctoral research at EPFL, the Swiss Federal Institute of Technology in Lausanne. He also conducted research at MIT’s Computer Science and Artificial Intelligence Laboratory.

Stella co-founded Embodied AI with chief technology officer Kai Junge and chief operating officer Max Polzin, both of whom have doctoral backgrounds in robotics. Faber led the company’s first funding round, alongside TechShop, Look AI Ventures, Plug and Play and Vento.

For manufacturers, the next checkpoint is whether Air can move from testing into larger-scale production as Embodied AI targets CE marking in 2027. The roughly one-month learning cycle and contracts with major manufacturers could ease deployment, but the disclosed 98% reliability benchmark comes from a different robot.